Forums/GM Game 2018-19

2018-19 GM Game - Official Scoring Thread

Aug 21, 2018 at 9:12
#26
Thread Starter
Let'sGoBlues | WPGv4
Joined: Jun 2016
Posts: 4,269
Likes: 2,433
A: B. Gallagher: Actual - 41, Expected - 50
B: W. Karlsson: Actual - 66, Expected - 48

C: J. Schwartz: actual 62, expected 56

there are players that have good GF (or production) and good xGF. It's not grounding the results just because they dont perfectly align.
Aug 21, 2018 at 10:25
#27
rangersandislesfan
Joined: Mar 2017
Posts: 26,874
Likes: 2,908
Quoting: phillyjabroni
Any form of subjective rankings will be based on how well GMs perceive that team to be. Based on the rules doc, the way we are intended to vote on the teams is using in-real-life factors (fatigue, etc.).

If you'd rather use objective rankings (this model), that's perfectly fine. If you'd like to use subjective rankings (your personal evaluation of how well the team performs in the playoffs), then you're going to have to base your evaluation on whatever you use to evaluate them (eye test, stats, etc.)


I've always liked the idea of someone using the video game for this, but i have no idea if that's possible as i do not play the NHL video games.
Aug 21, 2018 at 10:32
#28
Stars GM for V3
Joined: Mar 2017
Posts: 7,029
Likes: 1,443
Quoting: A_K
Don't let the word 'expected' skew your interpretation of xGF. xGF% does measure the impact a player has right now. It uses real, measured shot impacts and says whether that player is producing quality chances at a good or bad ratio to what they're giving up. It can be used as a predictive stat but also as a descriptive one of past play. If we eliminate xGF% from the standings, we will lose that piece of the evaluation... pts (or +/-) won't show the entire impact that player is making on the game. That is why we proposed also tracking CF and xGF.

And if you're worried about losing credit due to differences in point production vs xGF%, keep in mind that the real production will be worth 3x more than the xGF%. It is proposed to be a small player in the grand scheme of things.


Perhaps I'm not being clear, but the amount of impact isn't the issue. I was using gf vs. Xgf for simplicity, but it might be simpler to state that shot quality doesn't matter if the player is not producing results. At least as it relates to position in the standings, which is our goal here.
If Ovechkin's goals for is 0 in 20 games, his team will drop in the standings.
If he has 60 goals for in the next 60 games, his team will rise in the standings.
If you incorporate his xGF (say 60), even if it's weighted lower, you are still buoying his team's positron in the standings by some amount through those 20 games. Instead of Washington dropping 10 spots through those 20 games, perhaps they only drop 8 or 9 places.

That impact would be magnified if you have multiple players that are "snake bitten". If I have 10 players underachieving, and being propped up by their xGF, Washington in the above example might only drop 3 or 4 spots in our standings.
Perhaps that isnt significant on a month to month basis if everyone regressed to the mean by the end of the season, but 1 or 2 places could make a difference in waiver order and final draft position, depending on the timing.

We WANT the standings to be re-ordered from scratch every month. Therefore it doesn't make sense to include expected numbers or "adjust" the actual performance of a player based on a predictive statistical model.
It's not who deserves to rank higher or owns the most players that are normally good, but who gets the best results from the players they own.

I hope that makes more sense. We can look into other alternatives. I realize WAR incorporates some predictive metrics, so isn't a perfect solution, but at least then something like xGF would be a fraction of a piece of the picture, as opposed to a full puzzle piece.
Aug 21, 2018 at 12:11
#29
NateElder12
Joined: Apr 2016
Posts: 5,736
Likes: 799
Edited Aug 21, 2018 at 12:22
Quoting: ricochetii
Perhaps I'm not being clear, but the amount of impact isn't the issue. I was using gf vs. Xgf for simplicity, but it might be simpler to state that shot quality doesn't matter if the player is not producing results. At least as it relates to position in the standings, which is our goal here.
If Ovechkin's goals for is 0 in 20 games, his team will drop in the standings.
If he has 60 goals for in the next 60 games, his team will rise in the standings.
If you incorporate his xGF (say 60), even if it's weighted lower, you are still buoying his team's position in the standings by some amount through those 20 games. Instead of Washington dropping 10 spots through those 20 games, perhaps they only drop 8 or 9 places.

That impact would be magnified if you have multiple players that are "snake bitten". If I have 10 players underachieving, and being propped up by their xGF, Washington in the above example might only drop 3 or 4 spots in our standings.
Perhaps that isnt significant on a month to month basis if everyone regressed to the mean by the end of the season, but 1 or 2 places could make a difference in waiver order and final draft position, depending on the timing.

We WANT the standings to be re-ordered from scratch every month. Therefore it doesn't make sense to include expected numbers or "adjust" the actual performance of a player based on a predictive statistical model.
It's not who deserves to rank higher or owns the most players that are normally good, but who gets the best results from the players they own.

I hope that makes more sense. We can look into other alternatives. I realize WAR incorporates some predictive metrics, so isn't a perfect solution, but at least then something like xGF would be a fraction of a piece of the picture, as opposed to a full puzzle piece.


If we can't agree on xGF% then I'm fine looking at other alternatives, but I am greatly against using +/- as a scoring method or just singular raw point totals only. All the good teams' players IRL will be the only ones in the game that has any value if that's the case. In the above example, you give about Ovi though that's not really what xGF% is showing. If Ovi has 0 goals in 20 games his team would not necessarily drop in the standings if he doesn't concede any goals in that 20 game stretch, and you would be able to get a better picture of his play using xGF%. If he's driving better scoring opportunities, but Tom Wilson just can't finish those chances, Ovi would then be punished in our game because his teammate is bad IRL? I'm not sure how that is any more fair, especially if in the game you surround the example player with better linemates than he has IRL. I believe that is why xGF% was added as such a small %. A % that would not greatly impact the standings but would be used as a neutral way to account for a large amount of team "shakeups" and identify which IRL players were driving positive results apart from just raw totals. A lot, if not most people, have been basing decisions regarding team building on how well each player they would bring in would play with the individuals on their fictitious teams - not IRL teams. If we are going off a system that only rewards IRL raw results without context that completely alters the market price of every individual in the game meaning if fantasy hockey style scoring is the route we are going I think that needs to be cleared up sooner rather than later. For me personally, I know I've targeted guys that don't necessarily win the scoring titles every year but drive positive results for their teammates and could exceed their IRL results if placed in positions to do so. Like I said, we don't have to use xGF%, but as a small % of the scoring system, it seemed like a neutral way to evaluate who had good seasons IRL among players who might not post 100+ point seasons consistently or even ever. Here are a few more things we could consider or look into, IMO, that others might like better (I'll provide last years "top-5 forwards min. 500 minutes TOI" to give everyone an understanding of who led in some categories, for reference):

GF%RelTM - goals for % relative to teammates (Wild Bill, Couturier, Faksa, Brown, and Matthews)
xGF%RelTM - expected goals for % relative to teammates (B. Ritchie, Lowry, Sedin, Couturier, Dadonov)
GS - Game Score (McDavid, Panarin, Kucherov, Pastrnak, Giroux)
GS/60 - Game Score per 60 (would need to set a minimum threshold of TOI for players to qualify though so this may be tricky) (Marchand, Bergeron, McDavid, C. Smith, Pastrnak)
SCF (on ice) - Scoring Chances For while on ice (McDavid, Crosby, Panarin, Schenn, Kucherov)
HDCF (on ice) - High Danger Chances For while on ice (McDavid, Crosby, Radulov, Gaudreau, Seguin)
SCF % (on ice)- Scoring Chances For % (Lowry, Schwartz, Heinen, B. Ritchie, Granlund)
HDCF % (on ice) - High Danger Chances For % (Lowry, Ritchie, Granlund, Koivu, Benn)
iSCF - Individual Scoring Chances For (McDavid, Skinner, Ovi, Tarasenko, MacKinnon)
iHDCF - Individual High Danger Chances For (McDavid, E. Kane, Hornqvist, Saad, Gallagher)

.. I know Rico suggested WAR, but most WAR/GAR models can't be done/calculated until the conclusion of the season leaving us in the dark until then (massive effects on TDL, waiver order, etc.). An alternative that can be calculated as the season progresses is Game Score (GS) - which I know AK you have mentioned you liked in the past. Maybe that is our middle ground? Idk, but these above suggestions are the ones I have that we can consider since nobody on either side is really suggesting anything else. I'd rather not debate about this for weeks and then the season is upon us if that makes sense. If nobody is satisfied with these please make suggestions or contribute instead of just taking one side or another. It will speed up the process instead of everyone in the league saying who they agree with and who they don't. This really shouldn't turn into a "pissing contest", but rather everyone finding a way to balance the scoring system to appropriately account/weigh each player (even the depth guys) on our teams to avoid a scoring system that only rewards the top-10 point scorers IRL.

cheers

Edit/P.S.:

.. I would like to add that you can see why it's important to add a context type stat that is weighted at least a little bit. B. Ritchie had a sneaky good year but nobody talks about him. He consistently was the leader on every line he was placed as shown from most of the stats above. It's important to note that he only had 80 minutes with Benn or Seguin and his positive numbers from xGF% should give him a boost in value since the majority of his minutes came with terrible teammates (Hanzal, Spezza, G. Smith, and Elie). This is why we need something to make up for this. If I traded for Ritchie and put him with say Pavs/Kreider there is no way he would have only scored 14 points all year based on those numbers.
phillyjabroni liked this.
Aug 21, 2018 at 4:00
#30
Quebec GM v4
Joined: Mar 2017
Posts: 9,276
Likes: 2,205
Quoting: rangersandislesfan
I've always liked the idea of someone using the video game for this, but i have no idea if that's possible as i do not play the NHL video games.


I've been in a similar situation where a video game was used to simulate the season and the results were completely unrealistic, even with adjusted overalls. It's not really a viable option for our purposes.
Pointerr21 liked this.
Aug 21, 2018 at 5:05
#31
Stars GM for V3
Joined: Mar 2017
Posts: 7,029
Likes: 1,443
Statistics which generally correlate with team success, which can also apply to players:

GF%: Shows what percentage of goals scored while that player is on the ice, are in that player's favor.

GF/60: Shows approximately how many goals a player scores per 60 minutes of play.

CF%: Shows a player's impact on shots for/against, and is generally a strong indicator of team success.

SH%: Shows a player's propensity for scoring with the shots they take.

SV%: Shows the percentage of shots faced which do not result in goals while a player is on the ice.

Theoretically, if you collect these stats for each individual player, weight those stats according to that player's TOI, combine the results from 23 players to get a "team" result, and then compare those numbers against like groups, you should get a realistic idea of how those teams would rank against each other.

If you use the expected values for those categories, you can get a decent indication of how those teams might rank against each other in the future.
If you combine both groups, you are diluting the goal.
For the regular season, the goal is to be current and compare those statistics with 31 other teams. Real life results make more sense.
For the playoffs, the goal is to predict the result of an imaginary confrontation between 2 teams. Expected or rated results make more sense (since we won't have real life results for a number of players).
Aug 21, 2018 at 6:43
#32
rangersandislesfan
Joined: Mar 2017
Posts: 26,874
Likes: 2,908
Quoting: phillyjabroni
I've been in a similar situation where a video game was used to simulate the season and the results were completely unrealistic, even with adjusted overalls. It's not really a viable option for our purposes.


I don't know what you mean by 'unrealistic' though.
Aug 21, 2018 at 6:47
#33
Joined: Jan 2017
Posts: 6,933
Likes: 1,339
Quoting: rangersandislesfan
I don't know what you mean by 'unrealistic' though.


A team with 12 4th liners can win the Stanley Cup, but a team with Sidney Crosby, Connor McDavid, John Tavares etc will come 31st.
Aug 21, 2018 at 6:52
#34
Thread Starter
Let'sGoBlues | WPGv4
Joined: Jun 2016
Posts: 4,269
Likes: 2,433
@ricochetii @NateElder12 @phillyjabroni I've been a little busy to make a quality post today but let's make sure the discussion is about xGF and not ixGF. Tracking individual goals for WOULD in fact dampen a player's overproduction (high sh%) whereas xGF would not necessarily.

And I want to make clear that xGF is predictive (future) AND descriptive (past and current).

"Expected Goals (xG) significantly outperforms score-adjusted Corsi (CF%) and Goals For (GF%) in predicting future goals at the team and player levels. xG is also descriptive, which makes it a superior tool in evaluating a team and player’s past and current offensive performance."

https://hockey-graphs.com/2015/10/01/expected-goals-are-a-better-predictor-of-future-scoring-than-corsi-goals/amp/?__twitter_impression=true

and I'm with Nate on GF%... it's basically +/-, I'd rather not include it for what we're trying to do.
phillyjabroni liked this.
Aug 21, 2018 at 7:09
#35
Stars GM for V3
Joined: Mar 2017
Posts: 7,029
Likes: 1,443
Quoting: A_K
@ricochetii @NateElder12 @phillyjabroni I've been a little busy to make a quality post today but let's make sure the discussion is about xGF and not ixGF. Tracking individual goals for WOULD in fact dampen a player's overproduction (high sh%) whereas xGF would not necessarily.

And I want to make clear that xGF is predictive (future) AND descriptive (past and current).

"Expected Goals (xG) significantly outperforms score-adjusted Corsi (CF%) and Goals For (GF%) in predicting future goals at the team and player levels. xG is also descriptive, which makes it a superior tool in evaluating a team and player’s past and current offensive performance."

https://hockey-graphs.com/2015/10/01/expected-goals-are-a-better-predictor-of-future-scoring-than-corsi-goals/amp/?__twitter_impression=true

and I'm with Nate on GF%... it's basically +/-, I'd rather not include it for what we're trying to do.


I seem to be the only one with these concerns, and I've tried to explain why in a number of ways, so if you still aren't convinced, I'm not going to continue debating and stand in the way of progress by myself.

GF% is indeed somewhat unreliable, as it shares many of the same pitfalls as +/-. It loses value when applied on an individual basis, but when used as a team metric it has value. Combining the individual numbers from a collection of players to acquire a "team" statistic is where I feel it would work.
If we focus on individual statistics that impact team performance, then work with those numbers to form a determination of what a team consisting of those 23 players would put up for the relevant numbers, I feel like it would be a better evaluation of a "team" than simply adding up a number of individual statistics, both real and projected.

Just to point out, even the article you shared is focused primarily on the merits of xGF as a predictor, not as a "real time" analytic.
Aug 21, 2018 at 7:57
#36
rangersandislesfan
Joined: Mar 2017
Posts: 26,874
Likes: 2,908
Quoting: jmac490
A team with 12 4th liners can win the Stanley Cup, but a team with Sidney Crosby, Connor McDavid, John Tavares etc will come 31st.


I'm talking about for playoffs. And in real life, any team in the playoffs can win the Stanley cup.
Aug 21, 2018 at 8:29
#37
Quebec GM v4
Joined: Mar 2017
Posts: 9,276
Likes: 2,205
Quoting: rangersandislesfan
I'm talking about for playoffs. And in real life, any team in the playoffs can win the Stanley cup.


To expand on "unrealistic" results, players who did well IRL these last couple of years were having results that were irregular. Guys like Jamie Benn, Sean Couturier, Auston Matthews, etc. had their production cut in half and had just awful results. On the flip side, guys like Mattias Janmark would have more than 70 points in the game.

It's not accurate enough to really consider using it for our purposes. It would be cool to see it, but the amount of time that needs to be allotted for overall adjustments and roster changes is an investment that has no return.
Aug 21, 2018 at 8:30
#38
GM - Avalanche
Joined: Dec 2016
Posts: 3,383
Likes: 677
Quoting: ricochetii
I seem to be the only one with these concerns, and I've tried to explain why in a number of ways, so if you still aren't convinced, I'm not going to continue debating and stand in the way of progress by myself.

GF% is indeed somewhat unreliable, as it shares many of the same pitfalls as +/-. It loses value when applied on an individual basis, but when used as a team metric it has value. Combining the individual numbers from a collection of players to acquire a "team" statistic is where I feel it would work.
If we focus on individual statistics that impact team performance, then work with those numbers to form a determination of what a team consisting of those 23 players would put up for the relevant numbers, I feel like it would be a better evaluation of a "team" than simply adding up a number of individual statistics, both real and projected.

Just to point out, even the article you shared is focused primarily on the merits of xGF as a predictor, not as a "real time" analytic.


No Ricochetti. I agree with your points. I just think xGF shouldn't be used.
ricochetii liked this.
Aug 21, 2018 at 9:46
#39
rangersandislesfan
Joined: Mar 2017
Posts: 26,874
Likes: 2,908
Quoting: phillyjabroni
To expand on "unrealistic" results, players who did well IRL these last couple of years were having results that were irregular. Guys like Jamie Benn, Sean Couturier, Auston Matthews, etc. had their production cut in half and had just awful results. On the flip side, guys like Mattias Janmark would have more than 70 points in the game.

It's not accurate enough to really consider using it for our purposes. It would be cool to see it, but the amount of time that needs to be allotted for overall adjustments and roster changes is an investment that has no return.


Maybe there should be a vote, but based on the votes for each team, they could have a random winner based on those votes. Example:

Seattle plays Houston in round one. Seattle gets 10 votes, and Houston gets 19. Then someone would use a random number generator to pick the winner, with Seattle getting a 10 in 29 chance of winning the series. Idk who would be allowed to vote, so it may be more than 29 voters ... but i think you get the idea.
Aug 21, 2018 at 10:02
#40
Blues GM v4
Joined: Nov 2016
Posts: 8,028
Likes: 2,407
Quoting: rangersandislesfan
Maybe there should be a vote, but based on the votes for each team, they could have a random winner based on those votes. Example:

Seattle plays Houston in round one. Seattle gets 10 votes, and Houston gets 19. Then someone would use a random number generator to pick the winner, with Seattle getting a 10 in 29 chance of winning the series. Idk who would be allowed to vote, so it may be more than 29 voters ... but i think you get the idea.


Why do this instead of statiscal measures that actually have justification?
NateElder12 liked this.
Aug 21, 2018 at 10:10
#41
@Pointerr21
Joined: Mar 2017
Posts: 7,432
Likes: 2,528
Quoting: rangersandislesfan
Maybe there should be a vote, but based on the votes for each team, they could have a random winner based on those votes. Example:

Seattle plays Houston in round one. Seattle gets 10 votes, and Houston gets 19. Then someone would use a random number generator to pick the winner, with Seattle getting a 10 in 29 chance of winning the series. Idk who would be allowed to vote, so it may be more than 29 voters ... but i think you get the idea.


How about just throw everyone's team name with the same odds into a hat and whoever's name gets pulled wins the 1st overall pick and the league?
NateElder12 liked this.
Aug 21, 2018 at 10:14
#42
rangersandislesfan
Joined: Mar 2017
Posts: 26,874
Likes: 2,908
Quoting: krakowitz
Why do this instead of statiscal measures that actually have justification?


That's fair, but how the team plays together is huge.
Aug 21, 2018 at 10:16
#43
Blues GM v4
Joined: Nov 2016
Posts: 8,028
Likes: 2,407
Quoting: rangersandislesfan
That's fair, but how the team plays together is huge.


While that is correct, how do you think your voting idea better accounts for that than data covering each player?
Aug 21, 2018 at 10:35
#44
rangersandislesfan
Joined: Mar 2017
Posts: 26,874
Likes: 2,908
Quoting: krakowitz
While that is correct, how do you think your voting idea better accounts for that than data covering each player?


Sorry, i got confused ... thought you were talking about the video game idea lol ... If the video game doesn't work then having a vote with a random number generator might be the best option, because we'd see more upsets ... if there's voting at all, i really hope people wouldn't just vote for whoever is better on paper ON EVERY SERIES. The problem is, even if one team gets some votes, the better team on paper will probably get more votes ... so if both teams in the series have a chance, that makes the most sense IMO.
Aug 21, 2018 at 10:54
#45
Thread Starter
Let'sGoBlues | WPGv4
Joined: Jun 2016
Posts: 4,269
Likes: 2,433
@ricochetii @TMLSage We (BOG) are discussing a fair path forward. I dont want you guys to feel unheard, but we are pretty confident in the model as it currently stands. It's worth noting that we included xGF as one of the ranking metrics in the original rules doc that everyone signed off on, but we always want to be democratic when issues arise. For now, we may just hold to see if there is dissent from more GMs.
Aug 21, 2018 at 11:00
#46
Blues GM v4
Joined: Nov 2016
Posts: 8,028
Likes: 2,407
Quoting: rangersandislesfan
Sorry, i got confused ... thought you were talking about the video game idea lol ... If the video game doesn't work then having a vote with a random number generator might be the best option, because we'd see more upsets ... if there's voting at all, i really hope people wouldn't just vote for whoever is better on paper ON EVERY SERIES. The problem is, even if one team gets some votes, the better team on paper will probably get more votes ... so if both teams in the series have a chance, that makes the most sense IMO.


But we play this entire game based off of making the best team on paper. So the best way to determine a winner is going to be who built the best performing roster on paper.
Aug 21, 2018 at 11:17
#47
rangersandislesfan
Joined: Mar 2017
Posts: 26,874
Likes: 2,908
Quoting: krakowitz
But we play this entire game based off of making the best team on paper. So the best way to determine a winner is going to be who built the best performing roster on paper.


But that isn't realistic ... and there is no 'winner of the GM Game'. There can be a cup winner though.
Aug 21, 2018 at 11:32
#48
@Pointerr21
Joined: Mar 2017
Posts: 7,432
Likes: 2,528
Quoting: rangersandislesfan
But that isn't realistic ... and there is no 'winner of the GM Game'. There can be a cup winner though.


You just contradicted yourself. You said there can't be a winner yet there can be a cup winner. A cup winner is a winner.
Aug 21, 2018 at 11:35
#49
@Pointerr21
Joined: Mar 2017
Posts: 7,432
Likes: 2,528
Quoting: rangersandislesfan
But that isn't realistic ... and there is no 'winner of the GM Game'. There can be a cup winner though.


And using analytics to determine a winner is the closest thing to "realistic."
Aug 21, 2018 at 11:51
#50
GM - Avalanche
Joined: Dec 2016
Posts: 3,383
Likes: 677
Quoting: BoltsPoint21
And using analytics to determine a winner is the closest thing to "realistic."


But some of the analytics such as xGF I'd argue are counter productive to describing the current season.
 
Reply
To create a post please Login or Register
Question:
Options:
Add Option
Remove Option
Submit Poll